In the spring of 2018, I presented at the Microsoft Data Science, Engineering, and Research Conference on a question that continues to shape the global technology landscape: Can I trust AI? Of the hundreds of proposals submitted that year, mine was the only one accepted from Microsoft’s legal department. Partnering with the well-respected Chief Architect - Data and AI, we simply explored: Can AI be trusted? The session detailed what companies like Microsoft were doing to earn that trust.

The session resonated. I was invited back that fall to lead a second discussion, this time alongside Microsoft’s Responsible AI team, focused specifically on ethical AI. I share this not as a credential, but as context. The question of trust in AI is not new. It is persistent, consequential, and increasingly urgent as AI systems grow more autonomous.

Today, the rapid rise of agentic AI (systems designed to operate with increasing independence, make decisions, and interact with other agents) has sparked renewed excitement across the technology ecosystem. Innovation is accelerating. Experimentation is encouraged. New platforms are rapidly emerging. Organizations across industries are exploring how these systems can enhance efficiency, insight, and scale.

Yet amid this momentum, one foundational question remains: Can I trust AI?

The answer is neither a simple “yes” nor a reflexive “no”. Trust in AI is directly proportional to trust in the governance frameworks that surround it. AI should only be trusted to the extent that its oversight mechanisms, accountability structures, risk mitigation protocols, and ethical guardrails are rigorously designed and consistently implemented - internally and externally. In many ways, a “zero trust” mindset is not skepticism; it is discipline.

This tension between innovation and safeguards is not new. In 2016, Microsoft launched Tay, an AI chatbot designed to learn conversational patterns through public interaction on Twitter (now X). Within hours, Tay was manipulated into producing offensive and harmful content, exposing how quickly machine learning systems can be exploited when governance mechanisms are not sufficiently mature.

Tay is often remembered as a failure, but the more instructive lesson, however, lies in Microsoft’s response. The system was taken offline. The shortcomings were studied. The learnings informed Microsoft’s later work on Responsible AI, embedding human oversight, risk assessment, and ethical design as foundational components rather than afterthoughts. The episode demonstrated that governance is not optional infrastructure; it is core architecture.

Several years later, generative AI entered the mainstream at unprecedented speed. When OpenAI’s ChatGPT launched publicly, it reached millions of users within days. Its swift adoption reflected overwhelming demand, but it also underscored a broader reality: innovation was moving faster than most regulatory, ethical, and enterprise governance frameworks were prepared to accommodate. Organizations found themselves building guardrails in real time, responding to active deployment rather than shaping it prior to scale. The technology did not wait for consensus. It moved forward, and governance efforts raced to keep pace.

Now as we examine the current evolution toward agentic AI…

New platforms such as Moltbook describe themselves as social environments where AI agents share, discuss, and upvote content while humans observe from the sidelines. At first glance, this may appear to be a natural extension of experimentation in artificial intelligence. But as systems begin interacting autonomously - not just with humans, but with one another - the complexity of oversight increases significantly.

Foundational questions demand attention:

Who is designing these agents?
What objectives are embedded within their architectures?
What incentives shape their behavior?
What escalation mechanisms exist when outputs produce unintended consequences?
Who is ultimately accountable?

Transparency is a cornerstone of trust. Understanding that the absence of clearly articulated governance systems creates risk, particularly before technologies reach scale is critical.

When systems are engineered to learn, adapt, and interact independently, accountability cannot be retrofitted after public exposure. History repeatedly demonstrates that governance implemented after harm occurs is far more costly than governance embedded from the outset.

Agentic AI introduces stakes that are even more complex. Autonomous agents have the capacity to inform, and in some contexts execute decisions across workflows, enterprises, and interconnected systems, often with limited real-time human intervention. As autonomous systems begin learning from one another, the velocity and scale of interaction increase exponentially. Without disciplined governance, risks compound just as quickly.

This is not an argument against agentic AI, and the concern is not about presuming malicious intent. However, sustainable innovation depends on intentional design, transparency, and disciplined oversight.

Trust is not granted by novelty or speed. It is earned through rigor, repeatability, and governance infrastructures that are as sophisticated as the systems they oversee.

If we expect autonomous systems to operate responsibly in the world, responsibility must be engineered into them from the beginning. Governance is not a barrier to innovation…it is what makes innovation durable.